Google is preparing to put its AI chips in orbit. Following years of research, the project is scheduled to send a prototype satellite into space to find out how the company's Tensor Processing Units hold up there. The test is the first orbital step for a project Google announced last year and describes as a long-term research effort into whether space could eventually host machine learning infrastructure at scale.
The appeal is sunlight. In low Earth orbit, satellites can soak up near-constant solar energy, which Google says works out to as much as eight times the solar power available on the ground. The company's longer-term idea is to network multiple constellations of satellites so they can handle bigger AI workloads while in orbit, though nothing about that is being tested yet.
Google frames the launch as an experiment rather than a product. The company compares the work to early research into autonomous driving and quantum computing, saying both needed years of experimentation before they became practical systems. That is the company's stated reason for moving in small, deliberate steps rather than promising orbital data centers now.
The announcement does not include a launch date, a launch provider, a cost, or the size of the satellite, and it does not say how long the test is expected to run. It also does not address how data would travel between orbiting chips and the ground, or what happens to hardware that cannot be repaired once it is up there.
The orbital test extends Google's broader push to prove its TPUs can run AI workloads outside its own data centers. In June, the company backed a $3.2 billion data center project in western New York built around its TPUs, with capacity flowing through Fluidstack to Anthropic for training its Claude models. That arrangement was structured to lock in demand for Google's chips rather than Nvidia's GPUs, mirroring how Nvidia has long helped customers finance large clusters.
Google's TPUs started out running internal services such as Search and YouTube before the company began pitching them as an alternative to Nvidia's hardware in the AI training market. Suncatcher takes that pitch somewhere no data center can go, though the company is careful to call it research rather than a deployment plan.
What the first flight can realistically show is narrow: whether TPUs function in the radiation and temperature swings of orbit, and whether a satellite can generate enough power to run them. Google says the goal behind the work is making AI's benefits in areas such as healthcare and scientific discovery reach everyone far into the future. Whether that requires chips above the atmosphere is exactly what the prototype is meant to start answering.
